Percolation analysis 1 (the pattern)
Back in March I reported on the set of 1000 most faved photos, posted some statistics about them, and hinted at a more interesting analysis I had done. I then presented that analysis in a couple of talks, including at Flickr HQ, but never got around to posting it online.
To recap a bit, I performed rather extensive (though by no means comprehensive) crawling of flickr's database, collecting information about 15000 highly-faved images, the list of favorites for a very large number of flickr members, and more contact list information that one can try to comprehend.
Using these and other types of information from flickr, one can do all kinds of analyses, for example:
• Using information about image context (photoX was added to set1, photoY was sent to groupA), one can do photostream visualizations.
• Using information about group contents (photoX was sent to groupA) and some trickery, one can create a network visualizing the relationships between similar groups and even have some fun browsing related groups.
Fun as they can be, all these analyses have one thing in common: each is based on a single type of information. For some time, I tried to think what interesting results could be obtained from combining several information types. The analysis I'll be showing here does precisely that.
As you can see in the image, I define a basic unit of information, a basic pattern, that encapsulates the following:
1) A user ("Owner") posted a photo. [red]
2) Another user ("Viewer") added a third user ("contact") to his/her social network. [green]
3) The third user ("contact") marked the photo as a favorite. [pink]
As you can see, I'm integrating into a basic pattern three types of information, represented by the red, green and pink arrows.
(See the next slide...)